[R-pkg-devel] Any new developments on "apparent S3 method" issue?

[Uwe Ligges]

On 04.05.2017 16:23, Lenth, Russell V wrote:
> OK thanks -- this gives me a new perspective. I see this result:
>
>     > tools::nonS3methods("coda")
>     [1] "as.mcmc.list"
>
> which indicates more about the mechanisms in place for this kind of situation. So now it seems that if package 'pkg_B' simply imports and exports a function 'looks.like.S3.method'  from package 'pkg_A', and 'looks.like.S3.method' is already known via tools::nonS3methods("pkg_A") and thus has a check exclusion, then 'pkg_B' should have the same check exclusion. I recognize that such a check-exclusion-inheritance provision may not be very straightforward to implement.

Ideally, pkg_B avoids the infelicities that had been established with pkg A.

Best,
Uwe Ligges



>
> Russ
>
> -----Original Message-----
> From: Uwe Ligges [mailto:ligges at statistik.tu-dortmund.de]
> Sent: Thursday, May 4, 2017 3:06 AM
> To: Lenth, Russell V <russell-lenth at uiowa.edu>; r-package-devel at r-project.org
> Subject: Re: [R-pkg-devel] Any new developments on "apparent S3 method" issue?
>
>
>
> On 03.05.2017 18:39, Lenth, Russell V wrote:
>> Package developers list:
>>
>> I recently added an 'as.mcmc.list' method for objects in one of my packages. Now, in doing an --as-cran check, I get this message:
>>
>> * checking S3 generic/method consistency ... NOTE Found the following
>> apparent S3 methods exported but not registered:
>>   as.mcmc.list
>> See section 'Registering S3 methods' in the 'Writing R Extensions' manual.
>>
>> Apparently, the checking routines conclude that I should register an 'as.mcmc' method for class "list" -- something that does not exist and is not needed. I did some searching, and found an exchange on this issue from January 2015, in which a "nonS3Method()" construct was proposed. That sure looks like what is needed, but it still doesn't exist. Is there a different way to handle this that has been developed?
>>
>> Further information -- this is related to the *coda* package, and my NAMESPACE file contains this code:
>>
>> if (requireNamespace("coda", quietly = TRUE)) {
>>     importFrom(coda, as.mcmc, as.mcmc.list)
>>     export(as.mcmc, as.mcmc.list)
>>     S3method(as.mcmc, ref.grid)
>>     S3method(as.mcmc.list, ref.grid)
>> }
>>
>> This seems clear enough; and what's mysterious is that *coda* passes the CRAN checks without this same NOTE.
>
> Yes. coda had the name for a very long time and it was not reasonable to fix this naming in felicity later on, hence there is a check exclusion in R-devel/src/library/tools/R/utils.R for as.mcmc.list in package coda.
>
> Best,
> Uwe Ligges
>
>
>
>
>>
>> As a possible workaround, I am considering:
>>    -- providing and registering my own S3 method 'as.mcmc' for class "list" --- in spite of my disinterest in providing this
>>   -- not importing coda::as.mcmc.list and instead exporting my own
>> generic that handles objects of class "list" separately
>>
>> Any suggestions?
>>
>> Russ
>>
>> Russell V. Lenth  -  Professor Emeritus Department of Statistics and
>> Actuarial Science The University of Iowa  -  Iowa City, IA 52242  USA
>> Voice (319)335-0712 (Dept. office)  -  FAX (319)335-3017
>>
>> ______________________________________________
>> R-package-devel at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-package-devel
>>